PERSON DIRECTORY
Mark
Mark appears in 1 indexed conversation across 硅谷101. This directory brings every appearance, source, TL;DR, digest, and transcript into one searchable feed.
1 EPISODE1 SHOW
Language
E251|推理芯片之战:聊聊Groq、Cerebras与OpenAI三大路径与Bill Dally的设计哲学
推理竞争正从峰值算力转向每百万Token成本,decode反复读取全模型,使带宽成为核心瓶颈,SRAM的带宽经济性因此显现。Groq依靠编译期调度,Cerebras依靠晶圆级集成,但MoE路由、通信、良率、供应链与动态增长的KV cache仍是规模化验证的关键风险。
